Shangtang (00020) 2026 Interim Report: Gross Profit Growth Accelerates, Business Model Shifts, and Profit Quality Improves

On the afternoon of August 26, an interim report was updated on the Hong Kong Stock Exchange bulletin board, and the answer that the market had been waiting for four and a half years was finally revealed.

Shangtang Technology (00020) achieved IFRS net profit of 617 million yuan in the first half of 2026. This is the first time since Shangtang was listed in December 2021 that it has recorded a consolidated profit under IFRS.

The numbers themselves are shocking enough. But what really breathed global investors was the looming valuation anchor migration curve behind this financial report — from “leading technology” to “commercial execution,” Shang Tang is rewriting the pricing logic of the AI sector of Hong Kong stocks.

Profit results: Let's first see how the 617 million dollars came about

Profit is accounted for; recurring revenue (RR) and gross margin are the only business. To interpret Shangtang's 2026 interim report, let's first take a look at how the 617 million profit came from.

On the one hand, it is a substantial improvement in the main business. In the first half of the year, Shangtang achieved revenue of 2,911 billion yuan, up 23.4% year on year; gross profit of 1.26 billion yuan, up 32.9% year on year; gross margin reached 41.4%, up 2.9 percentage points year on year. Revenue growth outperformed cost growth, and scale effects began to unleash. At the same time, R&D expenditure was 1,756 billion yuan, a year-on-year decrease of 17.1%; administrative expenditure was 518 million yuan, a decrease of 13.2% year-on-year. Between one liter and one drop, profit margins are opened in both directions.

On the other hand, it is the fair value of ecological investment under the “1+X” strategy. Other net gains in the first half of the year reached $2,127 million, mainly including profits of $521 million from the sale of subsidiaries and unrealized gains of $1,249 million on financial assets measured at fair value. This portion of revenue is a capitalized reflection of Shangtang's forward-looking layout of ecological companies such as Daxiao Robotics and Shangtang Medical over the years.

Therefore, Shangtang's book profit comes from two lines. One is a sharp decrease in main business losses over the same period last year, and the other is fair value change income from AI ecosystem enterprise investments. The former measures the slope of management improvement, while the latter comes from phased financial releases brought about by the increase in ecological enterprise value under the “1+X” strategy, including profits from the sale of subsidiaries and changes in the fair value of financial assets. It is a phased monetization of many years of technology and industrial layout.

The mixed discussion led to a misjudgment. If you look at it separately, it is more clear: main business reduced losses by 67.3%, gross margin increased 2.9 percentage points to 41.4%, and RR surged 124% year over year. The three indicators improved simultaneously, indicating that the business model is shifting gears and the quality of profit is improving.

Business model: RR was first disclosed, and revenue moved from “project” to “service”

Shang Tang disclosed recurring revenue (RR) for the first time. The RR for the first half of 2026 was 1.145 billion yuan, up 124.4% from 510 million yuan in the first half of 2025, and the share of the group's revenue rose from 21.6% to 39.3%.

RR refers to revenue based on contracts already in force during the reporting period with continuous renewal attributes. Its rapid rise means that cooperation between Shangtang and customers is shifting from single project delivery to continuous service, and from “selling systems” to “selling subscriptions.” For an AI company that is still in the investment phase, the RR share jumped from 20% to nearly 40% within half a year, which explains the problem better than the profit change itself.

The revenue structure is also shifting gears. Generative AI revenue was 2,327 billion yuan, up 28.2% year on year, accounting for nearly 80% of the group's revenue. It is already the main engine; visual AI revenue was 497 million yuan, up 13.9% year on year, and growth resumed after optimizing business strategies. The revenue contribution rate of old customers was 67%, and continued to act as the company's entry point into the industry and overseas markets. Overseas business revenue increased 127.0% year on year, far higher than the Group's overall growth rate of 23.4%. Relying on mature visual AI capabilities, unified multi-modal models and localized delivery systems, the capacity is expanding outward.

The full picture of the shift can be summed up in a few figures: gross margin increased from 38.5% to 41.4%, R&D expenses fell 17.1% year over year to 1,756 billion yuan, administrative expenses fell 13.2%, and revenue was still growing at a rate of 23.4%. Revenue growth outperformed cost growth, and scale effects began to unleash.

Why can these three metrics improve at the same time

The three indicators improved at the same time, corresponding to the system-level capabilities of Shang Tang's “one model, one token factory, and one smart device management and control system” that began to run smoothly. The three parts are interlocking.

Let's look at the model first. The Neo-Unify architecture was proposed in March 2026, the SenseNova U1 with native and unified multi-modal understanding and generation was released in April, the SenseNova U1 Pro with GPT-Image 2 was launched at the World Artificial Intelligence Conference in July, and the open source SenseNova U1.5 Lite supports native 4K output with lightweight specifications in August. In terms of long-range agents, SenseNova 6.7 Flash Lite, which was released in May, has capabilities such as tool orchestration and long-term sequential memory. In scenarios such as information search, token consumption is reduced by 60% compared to plain text agents. Native multimodality determines what an intelligent body can understand and create, and a long-range agent determines whether it can continue to act and achieve goals. The two routes are integrated, and AI expands from a single point of capability to cover the entire chain of understanding, planning, execution, verification, and delivery.

Let's take another look at the Token Factory. The average daily token service volume exceeded 2.4 trillion in July 2026, an increase of about 22 times over the previous year. The large device has been upgraded from self-use computing power to a production token factory to provide services to 4 external basic model manufacturers. The agent model context training speed has been increased by 2 times, and the mainstream domestic chip MFU can reach up to 2.5 times the original chip manufacturer's benchmark level. The computational power collaborative agent achieved a 96% load prediction accuracy rate. The first half of the year saved more than 12 million yuan in electricity costs, reduced carbon emissions by 24,000 tons per 10,000 P of computing power, and continued to reduce the unit intelligent cost.

Finally, let's take a look at the intelligent control system. In the first half of the year, it served more than 1,000 enterprise customers and added more than 100 companies, covering more than 20 industries; Little Raccoon already served leading companies such as Lenovo, Ping An Technology, the three major telecom operators, and JD; Seko produced 10,000 minutes of video in a single day, with a cumulative estimated number of views exceeding 1.5 billion; and the cumulative number of users of the Personal Assistant Kappi series exceeded 45 million. Front-end products are diverse, back-office capabilities are unified, and the same set of task delivery capabilities can be quickly reused in different industries.

The logic is simple: customers and tasks continuously generate data and feedback to drive model, workflow and product iteration; improve model capabilities and token production efficiency, reduce task delivery costs, and expand application boundaries. Value measurement gradually moved from token consumption to task delivery, and a positive cycle between technical capability, customer value, recurring revenue, and operational efficiency changed.

Valuation anchor migration: from “technology leading” to “market leading”

The significance of the profit announcement is not that the 617 million figure itself is that it confirms the migration of a valuation anchor. Shang Tang is shifting from a model company that “sells APIs and computing power” to a service company that “delivers the ability to deliver trusted multi-modal intelligence.” The valuation anchor shifts from model ability to commercial profitability; this is the part that the market should really reevaluate.

The brokerage firm is already adjusting pricing. Goldman Sachs later released a research report on August 17, raising the target price of Shangtang from HK$2.00 to HK$2.05, but drastically adjusted the profit forecast: the net loss forecast for 2026 was narrowed from $1.25 billion to $59 million, and narrowed from $357 million to $336 million in 2027, which meant that Shangtang was close to the break-even point; profit forecasts for 2028 to 2032 were raised by 13%, 5%, 2%, 2%, and 1%, respectively.

Goldman Sachs judged that Shang Tang's AI model is moving towards “delegated intelligence” (Delegated Intelligence). In the future, users will focus on the delivery of AI output results, which will help large-scale commercialization. According to broader sellers' consensus, Shangtang's target price range for the past 12 months fell between HK$2.20 and HK$2.72, with an average value of around HK$2.43. There is still significant room for improvement from the current price of HK$1.47.

Support at the financial and index levels is also accumulating. Shang Tang has been included in core indices such as MSCI China and Hang Seng Technology. Passive allocation requirements are superimposed with active revaluation logic. And when an AI company proves that it can achieve positive operating profits, the valuation anchor point will naturally shift from “market sales ratio” to “enterprise EV/EBITDA” and “price-earnings ratio (PE)” — Shang Tang is at the critical point of this transition.

Conclusions

Initial profit is the starting line for valuation anchor migration.

At the same time, an AI company reduced losses by 67%, gross margin increased by 41%, and RR doubled year-on-year in the first half of the year. The signal conveyed was no longer “whether it can make money”, but “the way to make money is changing”: from one-time project delivery to continuous service, from selling tokens to delivering an intelligent body that can work.

The revaluation of Shang Tang's valuation has just begun.